The detection of bugs in software is divided into two research fields. Static analysis report warnings at the line level and are often false positives. Statistical models use historical change measures to predict bugs in commits at a higher level. We developed a tool which combines both of these approaches.

Our tool analyses each commit of a project and identifies which commit introduced a warning. It processed over 45k commits, more then previous research. We propose an augmented bug model which includes static analysis measures which found that a twofold increase in the number of new warnings increases the odds of introducing a bug 1.5 times. Overall, our model accounts for 22% of the deviance which is an improvement over the 19.5% baseline. We demonstrate that we can use simple measure to predict new security warnings with a deviance explained of 30% and that recent development experience and more co-developers reduces the number of security warnings by 8%.

We perform a user study of developers who introduced new warnings in 37 projects. We found that 53% and 21% of warnings in Findbugs and Jlint respectively are useful. We analysed the time delta between the introduction and response of the developer to the notification of the warning. We hypothise that remembering the context  of the change as an impact on the perceived usefulness given useful warnings had a median of 11.5 versus 23 days for non useful warnings